Abstract: Hepatocellular carcinoma (HCC) is a primary liver malignant tumor with a high recurrence rate and poor prognosis, even after resection or liver transplantation, there is still a great possibility of recurrence. The accurate identification of the histological grade of hepatocellular carcinoma plays a vital role in the choice of the patient's treatment. In this paper, we discussed the histological diagnosis model of hepatocellular carcinoma based on deep learning methods and the clinical diagnosis experience of radiologists. We proposed the "self-attention" scheme of the dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) images and the corresponding model which can calculate and utilize the "self-attention" weights for the HCC histological grade classification task. The innovation of this method lies in calculating the importance of each sequence and each channel for the final classification task and combining the calculated weights with the spatiotemporal information contained in the enhanced sequence which can improve the classification performance. We conduct experiments on the self-built dataset based on clinical data. The experimental results show that the classification performance of our proposed "self-attention" model is better than that of the baseline and benchmark models in both the WHO and the Edmonson HCC histological grading standard. The classification accuracy of the "self-attention" model reaches 80%, the sensitivity is 82%, and the accuracy is 82%.